Biomedical Engineering Reference
In-Depth Information
be very time consuming, and important data can be lost during the transfer.
For example, information in the image header about whether the patient is
prone or supine, or head first or feet first must be accurately transferred along
with the image pixel values, if the patient's orientation within the scanner is
to be known. Errors in this process can result in the left and right sides of the
patient being flipped, which is potentially disastrous. The topic of data trans-
fer is treated in more detail in Chapter 4.
10.2.2
Registration Methods
Many algorithms have been proposed for registration of MR and CT images
of the head. As described in Chapter 3, these methods can be categorized into
those that make use of geometrical features in the images (such as points or
surfaces) and those that make use of voxel intensity values.
10.2.2.1
Registration Using Geometrical Features
The earliest techniques for registration of MR and CT images of the head
were use of point landmarks
1
2,6,7
Point landmarks can
either be point-like anatomical structures within the images visible in both
modalities
or stereotactic frames.
1,8
or they can be external fiducial markers rigidly affixed to the
9
10-12
skull
Whether the points are anatomical features
or external markers, the most common registration approach is to find the
rigid-body transformation that aligns the points in the least-squares sense,
as described in Chapter 3. When using external fiducial markers, the mark-
ers need to be visible in both modalities and must not introduce artifacts in
the images. Fatty markers appear very bright in MR, but are a poor choice
because protons in fat have a different resonant frequency from protons in
water; therefore they appear displaced (in the readout direction) relative to
their true position as compared to soft tissues of interest. The amount of
displacement depends on the readout gradient strength, as discussed in
Chapter 5. A mixture of MR contrast material and CT contrast material is a
better choice,
or attached to the skin.
9
but care must be taken to ensure that the markers are visible
in all MR sequences of interest. Certain types of inversion recovery
sequences can make features with particular relaxation times virtually invis-
ible in the images, and if the markers have this relaxation time, they will not
be visible in the images.
External fiducial markers rigidly attached to the skull have considerable
advantages over point-like anatomical features for registration. First, they
can appear sufficiently bright in the images to be easily identified by a user
or computer algorithm. Second, it is possible to calculate the position of
these features with an accuracy better than the voxel dimensions by using
a center of gravity calculation,
13
provided the markers are large enough to
appear in multiple voxels in all dimensions. Skin-attached markers can be
identified accurately in the images, but they can move relative to features
of interest in the patient, which degrades the accuracy. The identification of
